VP of Software Development Will Anderson lives his passion, which is programming. He started dabbling in computers in high school and turned his favorite hobby—writing software—into a successful career. In this Q&A, Will explains how Conexiom uses ML and AI.
Describe your role as VP of software development within the Conexiom team.
I split my time between planning for short- and mid-term, knowing where we want to take the product and planning ahead to get there.
The other part of my role is actually landing that plan. We do quarterly releases and midstream releases as needed. Making sure those releases happen is something we’re mindful of and dedicate time and effort toward achieving.
What about working at Conexiom do you most enjoy?
We make an exciting product. It’s not just another software package; we’re doing something that has not been done before: applying machine learning (ML) to the enterprise space for sales order automation.
We have an opportunity to change the market by building an enterprise-grade automation solution that the market hasn’t seen before. Our platform is purpose built. We are problems solvers. Delivering innovation like ours is a challenge to build, but we deliver on that promise. Now we need to get that message out. The sky’s the limit for our technology and our customer’s success.
“We have an opportunity to change the market by building an enterprise-grade automation solution that the market hasn’t seen before.”
How are ML and AI implemented into Conexiom?
The industry is still coming to terms with clear definitions of artificial intelligence (AI) and machine learning (ML). It’s quickly evolving out of the academic and research realms and into the business and enterprise software space. Conexiom is building machine models based on an extremely large data set. Because Conexiom processes more than 1 million documents each month, we have a rich history of data. And that’s how ML improves and is able to solve any document we throw at it.
In developing each change, we look at new algorithms or changes to existing ones and check them across the body of example documents to see how effective a particular change is at improving our system. The best changes make it into the next release.
Where AI differs from ML is AI is the intelligence piece. We have humans interpret and guide the ML implementation. But a true learning system, a self–learning system—that’s AI—where the AI is making informed decisions and recommendations. Therefore, it can make decisions and iterate through changes faster, disregarding the changes that do not work and building and improving on the changes that do work. This increases the overall rate of improvements. And this is where we are breaking new ground.
How does ML influence Conexiom’s product roadmap?
We are putting a lot of energy into the building and research of our ML modules. We already have many in the market. We initially conquered the low–hanging fruit. Now, we’re getting to more mature and complex ML models.
The customer sees it when they submit, for example, a new document format that’s never been seen before by our team or software. Instead of throwing people at that new document to help understand what the different elements are and what they mean, Conexiom uses a contextual transformation process to determine which elements are where and present that information back to the customer.
Conexiom’s Q4 2020 release incorporates ML functionality, particularly, Base Automation, which is the new capability to perform document transformation without having a pre-made configuration, which we use for touchless processing.
Why is 100% data accuracy and touchless automation a gold standard for manufacturers and distributors?
It comes down to quality. In your customer’s eyes, you are only as good as your last processed order. If we were to put through an order that was not 100% accurate, our customers couldn’t trust the technology.
“Some of our customers have a sophisticated automated distribution. They go from a customer order that processes through Conexiom all the way to picking and packing in 15 minutes. So the accuracy is what’s important. If you have people reviewing orders, you lose the value of automation. 100% accuracy is something we focus on and hold ourselves accountable for.”
So how does Conexiom extract data with 100% accuracy?
Conexiom deals with the actual text in incoming orders, as opposed to an optical character recognition (OCR) solution, which uses pictures of letters and numbers. We process the actual data that’s in the document. It starts with the correct data, and then we add the transformations and business-specific rules to that.
For example, a part number is typically an 8-, 10–, or 12–plus character value, usually numbers and letters. Because we start with the actual text, we know that we have the correct part number. And then we can implement business rules around quantity, price, or required products.
How does Conexiom process orders in various languages?
The software has been architected for different languages. If you think of an incoming document format in English, one customer may use the term “delivery method,” while another customer might use “DEL,” or “DL.” So the software allows for various labels to find the value for “delivery method.”
If that label is in a different language, it’s no different than the many variations just described. That’s how Conexiom processes documents in many languages. Everything is built off a relative reference label.
Why is sales order automation an easy solution to implement enterprise-wide?
Here’s where Conexiom really shines. Traditionally, an automation value would go through two IT projects: The buyer and the supplier both have to work out the technical details to make an automation project work. This is usually electronic data interchange (EDI). Conexiom, on the other hand, takes the incoming order from the buyer and does all the heavy lifting to deliver that document into the enterprise resource planning (ERP) system; the buyer doesn’t have to lift a finger.
We do that without asking questions about the technical fields and the difficulty around them. That’s why we can onboard clients so quickly.
We also implement business rules—such as minimum quantities, price discounts, part number, or address lookups—things that are specific to the business. We implement that business–level conversation in the software for a custom configuration for each customer.
Conexiom uses proprietary algorithms in its technology. Where did those algorithms come from?
Our proprietary algorithms were originally built on the sum knowledge of people that were manually doing custom programming, and then we built a system to do it for us.
Now, we’re seeing the non-obvious custom algorithms in the software. So by looking at the data, the system is adapting and learning. That’s where we’ll bridge into AI. But the sum of our algorithms is greater than the number of algorithms and business ideas that we first went into it with. So the system is basically teaching us.
We’re in the season of new-year resolutions. Do you have any 2021 resolutions for Conexiom?
Yes, it’s about investing more in our ML. That means thinking about how we can approach a problem, not starting with what we think the solution is. And having that faith to try something that maybe won’t work. And then we’ll find what does work. Once we define what works, we wrap them up and put them into the product.
Have you written software that you use in your personal life?
I’ve written fun things for myself around home- and project-planning. I approach it more technically than the average person. I’ve got a spreadsheet that can model pretty much anything.
If there was one manual process that you could replace with technology, what would it be?
I would use AI for groceries. I could tell it what I want to eat, and it’ll figure out what groceries I need. It would check for the ingredients I do or don’t have, what’s expired, and have food delivered so I can make meals each week.
Hear More Conexiom Voices, including how Conexiom delivers 100% data accuracy, straight from the EVP of product management.